bsd/sys/mcache.h

   1 /*
   2  * Copyright (c) 2006-2012 Apple Inc. All rights reserved.
   3  *
   4  * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
   5  *
   6  * This file contains Original Code and/or Modifications of Original Code
   7  * as defined in and that are subject to the Apple Public Source License
   8  * Version 2.0 (the 'License'). You may not use this file except in
   9  * compliance with the License. The rights granted to you under the License
  10  * may not be used to create, or enable the creation or redistribution of,
  11  * unlawful or unlicensed copies of an Apple operating system, or to
  12  * circumvent, violate, or enable the circumvention or violation of, any
  13  * terms of an Apple operating system software license agreement.
  14  *
  15  * Please obtain a copy of the License at
  16  * http://www.opensource.apple.com/apsl/ and read it before using this file.
  17  *
  18  * The Original Code and all software distributed under the License are
  19  * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
  20  * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
  21  * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
  22  * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
  23  * Please see the License for the specific language governing rights and
  24  * limitations under the License.
  25  *
  26  * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
  27  */
  28 #ifndef _SYS_MCACHE_H
  29 #define _SYS_MCACHE_H
  30
  31 #ifdef KERNEL_PRIVATE
  32
  33 #ifdef  __cplusplus
  34 extern "C" {
  35 #endif
  36
  37 #include <sys/types.h>
  38 #include <sys/queue.h>
  39 #include <mach/boolean.h>
  40 #include <kern/locks.h>
  41 #include <libkern/OSAtomic.h>
  42
  43 #ifdef ASSERT
  44 #undef ASSERT
  45 #endif
  46
  47 #ifdef VERIFY
  48 #undef VERIFY
  49 #endif
  50
  51 /*
  52  * Unlike VERIFY(), ASSERT() is evaluated only in DEBUG build.
  53  */
  54 #define VERIFY(EX)      ((void)((EX) || assfail(#EX, __FILE__, __LINE__)))
  55 #if DEBUG
  56 #define ASSERT(EX)      VERIFY(EX)
  57 #else
  58 #define ASSERT(EX)      ((void)0)
  59 #endif
  60
  61 /*
  62  * Compile time assert; this should be on its own someday.
  63  */
  64 #define _CASSERT(x)     \
  65         switch (0) { case 0: case (x): ; }
  66
  67 /*
  68  * Atomic macros; these should be on their own someday.
  69  */
  70 #define atomic_add_16_ov(a, n)                                          \
  71         ((u_int16_t) OSAddAtomic16(n, (volatile SInt16 *)a))
  72
  73 #define atomic_add_16(a, n)                                             \
  74         ((void) atomic_add_16_ov(a, n))
  75
  76 #define atomic_add_32_ov(a, n)                                          \
  77         ((u_int32_t) OSAddAtomic(n, (volatile SInt32 *)a))
  78
  79 #define atomic_add_32(a, n)                                             \
  80         ((void) atomic_add_32_ov(a, n))
  81
  82 #define atomic_add_64_ov(a, n)                                          \
  83         ((u_int64_t) OSAddAtomic64(n, (volatile SInt64 *)a))
  84
  85 #define atomic_add_64(a, n)                                             \
  86         ((void) atomic_add_64_ov(a, n))
  87
  88 #define atomic_set_64(a, n) do {                                        \
  89         while (!OSCompareAndSwap64(*a, n, (volatile UInt64 *)a))        \
  90                 ;                                                       \
  91 } while (0)
  92
  93 #if defined(__LP64__)
  94 #define atomic_get_64(n, a) do {                                        \
  95         (n) = *(a);                                                     \
  96 } while (0)
  97 #else
  98 #define atomic_get_64(n, a) do {                                        \
  99         (n) = atomic_add_64_ov(a, 0);                                   \
 100 } while (0)
 101 #endif /* __LP64__ */
 102
 103 #define CPU_CACHE_SIZE  64
 104
 105 #ifndef IS_P2ALIGNED
 106 #define IS_P2ALIGNED(v, a) \
 107         ((((uintptr_t)(v)) & ((uintptr_t)(a) - 1)) == 0)
 108 #endif /* IS_P2ALIGNED */
 109
 110 #ifndef P2ROUNDUP
 111 #define P2ROUNDUP(x, align) \
 112         (-(-((uintptr_t)(x)) & -(align)))
 113 #endif /* P2ROUNDUP */
 114
 115 #ifndef P2ROUNDDOWN
 116 #define P2ROUNDDOWN(x, align) \
 117         (((uintptr_t)(x)) & ~((uintptr_t)(align) - 1))
 118 #endif /* P2ROUNDDOWN */
 119
 120 #define MCACHE_FREE_PATTERN             0xdeadbeefdeadbeefULL
 121 #define MCACHE_UNINITIALIZED_PATTERN    0xbaddcafebaddcafeULL
 122
 123 /*
 124  * mcache allocation request flags.
 125  *
 126  * MCR_NOSLEEP and MCR_FAILOK are mutually exclusive.  The latter is used
 127  * by the mbuf allocator to handle the implementation of several caches that
 128  * involve multiple layers of mcache.  It implies a best effort blocking
 129  * allocation request; if the request cannot be satisfied, the caller will
 130  * be blocked until further notice, similar to MCR_SLEEP, except that upon
 131  * a wake up it will return immediately to the caller regardless of whether
 132  * the request can been fulfilled.
 133  *
 134  * MCR_TRYHARD implies a non-blocking allocation request, regardless of
 135  * whether MCR_NOSLEEP is set.  It informs the allocator that the request
 136  * should not cause the calling thread to block, and that it must have
 137  * exhausted all possible schemes to fulfill the request, including doing
 138  * reclaims and/or purges, before returning to the caller.
 139  *
 140  * Regular mcache clients should only use MCR_SLEEP or MCR_NOSLEEP.
 141  */
 142 #define MCR_SLEEP       0x0000          /* same as M_WAITOK */
 143 #define MCR_NOSLEEP     0x0001          /* same as M_NOWAIT */
 144 #define MCR_FAILOK      0x0100          /* private, for internal use only */
 145 #define MCR_TRYHARD     0x0200          /* private, for internal use only */
 146 #define MCR_USR1        0x1000          /* private, for internal use only */
 147
 148 #define MCR_NONBLOCKING (MCR_NOSLEEP | MCR_FAILOK | MCR_TRYHARD)
 149
 150 /*
 151  * Generic one-way linked list element structure.  This is used to handle
 152  * mcache_alloc_ext() requests in order to chain the allocated objects
 153  * together before returning them to the caller.
 154  */
 155 typedef struct mcache_obj {
 156         struct mcache_obj       *obj_next;
 157 } mcache_obj_t;
 158
 159 typedef struct mcache_bkt {
 160         void            *bkt_next;      /* next bucket in list */
 161         void            *bkt_obj[1];    /* one or more objects */
 162 } mcache_bkt_t;
 163
 164 typedef struct mcache_bktlist {
 165         mcache_bkt_t    *bl_list;       /* bucket list */
 166         u_int32_t       bl_total;       /* number of buckets */
 167         u_int32_t       bl_min;         /* min since last update */
 168         u_int32_t       bl_reaplimit;   /* max reapable buckets */
 169         u_int64_t       bl_alloc;       /* allocations from this list */
 170 } mcache_bktlist_t;
 171
 172 typedef struct mcache_bkttype {
 173         int             bt_bktsize;     /* bucket size (number of elements) */
 174         size_t          bt_minbuf;      /* all smaller buffers qualify */
 175         size_t          bt_maxbuf;      /* no larger bfufers qualify */
 176         struct mcache   *bt_cache;      /* bucket cache */
 177 } mcache_bkttype_t;
 178
 179 typedef struct mcache_cpu {
 180         decl_lck_mtx_data(, cc_lock);
 181         mcache_bkt_t    *cc_filled;     /* the currently filled bucket */
 182         mcache_bkt_t    *cc_pfilled;    /* the previously filled bucket */
 183         u_int64_t       cc_alloc;       /* allocations from this cpu */
 184         u_int64_t       cc_free;        /* frees to this cpu */
 185         int             cc_objs;        /* number of objects in filled bkt */
 186         int             cc_pobjs;       /* number of objects in previous bkt */
 187         int             cc_bktsize;     /* number of elements in a full bkt */
 188 } __attribute__((aligned(CPU_CACHE_SIZE), packed)) mcache_cpu_t;
 189
 190 typedef unsigned int (*mcache_allocfn_t)(void *, mcache_obj_t ***,
 191     unsigned int, int);
 192 typedef void (*mcache_freefn_t)(void *, mcache_obj_t *, boolean_t);
 193 typedef void (*mcache_auditfn_t)(void *, mcache_obj_t *, boolean_t);
 194 typedef void (*mcache_logfn_t)(u_int32_t, mcache_obj_t *, boolean_t);
 195 typedef void (*mcache_notifyfn_t)(void *, u_int32_t);
 196
 197 typedef struct mcache {
 198         /*
 199          * Cache properties
 200          */
 201         LIST_ENTRY(mcache) mc_list;     /* cache linkage */
 202         char            mc_name[32];    /* cache name */
 203         struct zone     *mc_slab_zone;  /* backend zone allocator */
 204         mcache_allocfn_t mc_slab_alloc; /* slab layer allocate callback */
 205         mcache_freefn_t mc_slab_free;   /* slab layer free callback */
 206         mcache_auditfn_t mc_slab_audit; /* slab layer audit callback */
 207         mcache_logfn_t mc_slab_log;     /* slab layer log callback */
 208         mcache_notifyfn_t mc_slab_notify; /* slab layer notify callback */
 209         void            *mc_private;    /* opaque arg to callbacks */
 210         size_t          mc_bufsize;     /* object size */
 211         size_t          mc_align;       /* object alignment */
 212         u_int32_t       mc_flags;       /* cache creation flags */
 213         u_int32_t       mc_purge_cnt;   /* # of purges requested by slab */
 214         u_int32_t       mc_enable_cnt;  /* # of reenables due to purges */
 215         u_int32_t       mc_waiter_cnt;  /* # of slab layer waiters */
 216         u_int32_t       mc_wretry_cnt;  /* # of wait retries */
 217         u_int32_t       mc_nwretry_cnt; /* # of no-wait retry attempts */
 218         u_int32_t       mc_nwfail_cnt;  /* # of no-wait retries that failed */
 219         decl_lck_mtx_data(, mc_sync_lock); /* protects purges and reenables */
 220         lck_attr_t      *mc_sync_lock_attr;
 221         lck_grp_t       *mc_sync_lock_grp;
 222         lck_grp_attr_t  *mc_sync_lock_grp_attr;
 223         /*
 224          * Keep CPU and buckets layers lock statistics separate.
 225          */
 226         lck_attr_t      *mc_cpu_lock_attr;
 227         lck_grp_t       *mc_cpu_lock_grp;
 228         lck_grp_attr_t  *mc_cpu_lock_grp_attr;
 229
 230         /*
 231          * Bucket layer common to all CPUs
 232          */
 233         decl_lck_mtx_data(, mc_bkt_lock);
 234         lck_attr_t      *mc_bkt_lock_attr;
 235         lck_grp_t       *mc_bkt_lock_grp;
 236         lck_grp_attr_t  *mc_bkt_lock_grp_attr;
 237         mcache_bkttype_t *cache_bkttype;        /* bucket type */
 238         mcache_bktlist_t mc_full;               /* full buckets */
 239         mcache_bktlist_t mc_empty;              /* empty buckets */
 240         size_t          mc_chunksize;           /* bufsize + alignment */
 241         u_int32_t       mc_bkt_contention;      /* lock contention count */
 242         u_int32_t       mc_bkt_contention_prev; /* previous snapshot */
 243
 244         /*
 245          * Per-CPU layer, aligned at cache line boundary
 246          */
 247         mcache_cpu_t    mc_cpu[1];
 248 } mcache_t;
 249
 250 #define MCACHE_ALIGN    8       /* default guaranteed alignment */
 251
 252 /* Valid values for mc_flags */
 253 #define MCF_VERIFY      0x00000001      /* enable verification */
 254 #define MCF_TRACE       0x00000002      /* enable transaction auditing */
 255 #define MCF_NOCPUCACHE  0x00000010      /* disable CPU layer caching */
 256 #define MCF_NOLEAKLOG   0x00000100      /* disable leak logging */
 257 #define MCF_EXPLEAKLOG  0x00000200      /* expose leak info to user space */
 258
 259 #define MCF_DEBUG       (MCF_VERIFY | MCF_TRACE)
 260 #define MCF_FLAGS_MASK  \
 261         (MCF_DEBUG | MCF_NOCPUCACHE | MCF_NOLEAKLOG | MCF_EXPLEAKLOG)
 262
 263 /* Valid values for notify callback */
 264 #define MCN_RETRYALLOC  0x00000001      /* Allocation should be retried */
 265
 266 #define MCACHE_STACK_DEPTH 16
 267
 268 typedef struct mcache_audit {
 269         struct mcache_audit *mca_next;  /* next audit struct */
 270         void            *mca_addr;      /* address of buffer */
 271         mcache_t        *mca_cache;     /* parent cache of the buffer */
 272         struct thread   *mca_thread;    /* thread doing transaction */
 273         struct thread   *mca_pthread;   /* previous transaction thread */
 274         size_t          mca_contents_size; /* size of contents */
 275         void            *mca_contents;  /* contents at last free */
 276         uint16_t        mca_depth;      /* pc stack depth */
 277         uint16_t        mca_pdepth;     /* previous transaction pc stack */
 278         void            *mca_stack[MCACHE_STACK_DEPTH];
 279         void            *mca_pstack[MCACHE_STACK_DEPTH];
 280         void            *mca_uptr;      /* user-specific pointer */
 281         uint32_t        mca_uflags;     /* user-specific flags */
 282 } mcache_audit_t;
 283
 284 __private_extern__ int assfail(const char *, const char *, int);
 285 __private_extern__ void mcache_init(void);
 286 __private_extern__ unsigned int mcache_getflags(void);
 287 __private_extern__ mcache_t *mcache_create(const char *, size_t,
 288     size_t, u_int32_t, int);
 289 __private_extern__ void *mcache_alloc(mcache_t *, int);
 290 __private_extern__ void mcache_free(mcache_t *, void *);
 291 __private_extern__ mcache_t *mcache_create_ext(const char *, size_t,
 292     mcache_allocfn_t, mcache_freefn_t, mcache_auditfn_t, mcache_logfn_t,
 293     mcache_notifyfn_t, void *, u_int32_t, int);
 294 __private_extern__ void mcache_destroy(mcache_t *);
 295 __private_extern__ unsigned int mcache_alloc_ext(mcache_t *, mcache_obj_t **,
 296     unsigned int, int);
 297 __private_extern__ void mcache_free_ext(mcache_t *, mcache_obj_t *);
 298 __private_extern__ void mcache_reap(void);
 299 __private_extern__ boolean_t mcache_purge_cache(mcache_t *);
 300 __private_extern__ void mcache_waiter_inc(mcache_t *);
 301 __private_extern__ void mcache_waiter_dec(mcache_t *);
 302 __private_extern__ boolean_t mcache_bkt_isempty(mcache_t *);
 303
 304 __private_extern__ void mcache_buffer_log(mcache_audit_t *, void *, mcache_t *);
 305 __private_extern__ void mcache_set_pattern(u_int64_t, void *, size_t);
 306 __private_extern__ void *mcache_verify_pattern(u_int64_t, void *, size_t);
 307 __private_extern__ void *mcache_verify_set_pattern(u_int64_t, u_int64_t,
 308     void *, size_t);
 309 __private_extern__ void mcache_audit_free_verify(mcache_audit_t *,
 310     void *, size_t, size_t);
 311 __private_extern__ void mcache_audit_free_verify_set(mcache_audit_t *,
 312     void *, size_t, size_t);
 313 __private_extern__ char *mcache_dump_mca(mcache_audit_t *);
 314 __private_extern__ void mcache_audit_panic(mcache_audit_t *, void *, size_t,
 315     int64_t, int64_t);
 316
 317 __private_extern__ mcache_t *mcache_audit_cache;
 318
 319 #ifdef  __cplusplus
 320 }
 321 #endif
 322
 323 #endif /* KERNEL_PRIVATE */
 324
 325 #endif /* _SYS_MCACHE_H */